New features are:

• Streamlined process + pipeline integration
• Wiktionary grounding + LLM explain
  • Enormous words data across multiple languages
  • Extremely detailed definitions
• New distribution format will be: jsonl, sqlite and more are to be determined
• Options are available to select specific category of words

Behold and stay tuned!

• Install project dependencies: uv sync
• Configure a .env file with DATABASE_URL
• Ensure a PostgreSQL database is reachable via that URL

Download the compressed dump:

uv run open-dictionary download --output data/raw-wiktextract-data.jsonl.gz

Extract the JSONL file:

uv run open-dictionary extract \
  --input data/raw-wiktextract-data.jsonl.gz \
  --output data/raw-wiktextract-data.jsonl

Stream the JSONL into PostgreSQL (dictionary_all.data is JSONB):

uv run open-dictionary load data/raw-wiktextract-data.jsonl \
  --table dictionary_all \
  --column data \
  --truncate

Run everything end-to-end with optional partitioning:

uv run open-dictionary pipeline \
  --workdir data \
  --table dictionary_all \
  --column data \
  --truncate

Split rows by language code into per-language tables when needed:

uv run open-dictionary partition \
  --table dictionary_all \
  --column data \
  --lang-field lang_code

Materialize a smaller set of languages into dedicated tables with a custom prefix:

uv run open-dictionary filter en zh \
  --table dictionary_all \
  --column data \
  --table-prefix dictionary_filtered

Pass all to emit every language into its own table:

uv run open-dictionary filter all --table dictionary_all --column data

Populate the common_score column with word frequency data (re-run with --recompute-existing to refresh scores):

uv run open-dictionary db-commonness --table dictionary_filtered_en

Normalize raw Wiktionary payloads into a slimmer JSONB column without invoking LLMs (writes to process by default):

uv run open-dictionary pre-process \
  --table dictionary_filtered_en \
  --source-column data \
  --target-column processed

Remove low-quality rows (zero common score, numeric tokens, legacy tags) directly in PostgreSQL:

uv run open-dictionary db-clean --table dictionary_filtered_en

Generate structured Chinese learner-friendly entries with the LLM define workflow (writes JSONB into new_speak by default). This streams rows in batches, dispatches up to 50 concurrent LLM calls with exponential-backoff retries, and resumes automatically on restart:

uv run open-dictionary llm-define \
  --table dictionary_filtered_en \
  --source-column data \
  --target-column new_speak

Provide LLM_MODEL, LLM_KEY, and LLM_API in your environment (e.g., .env) before running LLM commands.

Each command streams data in chunks to handle the 10M+ line dataset efficiently.